agentleFS
aaziraniaazirani/flutter_agentify/.agents/skills/edit-boundaries/SKILL.md

edit-boundaries

Skill0 starsChanged 4 days agoMITView on GitHub

What an agent may change freely, what needs the gate, what needs a question first and what is off-limits in Flutter Agentify. Use when a task touches dependencies, platform folders, CI, secrets, generated files or anything under packages/.

  • Reads credentials

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Don't load this for the order of work or the final report.

    line 8
  2. Distinctive:

    Before deleting data on the person's behalf, use the action with its approval step.

    line 40

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Edit boundaries

Don't load this for the order of work or the final report; see agent-workflow.

Rule

Sort every change into one of four groups before you make it.

  1. Free: the app's data, through actions. Creating, changing or deleting items and settings goes through registry.run(action, input) or dart run tool/action.dart <name> '<json>'. No gate is needed for a data change; actions that need approval ask the person.
  2. Gate: source code. Edits in lib/, test/, assets/lang/ and tool/ (except tool/upgrade.dart) are fine, and the work counts as done only after tool/check.sh ends with All checks passed.
  3. Ask first. New or upgraded dependencies in pubspec.yaml, platform configuration (android/, ios/, macos/, web/, windows/, linux/: permissions, entitlements, bundle ids, signing), CI files, and analysis_options.yaml. State what you want to change and why, then wait.
  4. Off-limits. Keys, tokens and .env files; generated files (*.g.dart, *.drift.dart, test/generated_migrations/); anything the upgrade owns (see below) unless the task is about the framework itself; raising a baseline in tool/baselines/ratchet.json.

Framework-owned paths change only with dart tool/upgrade.dart (see upgrade-the-framework):

  • packages/flutter_agentify_core, packages/flutter_agentify_ui, packages/flutter_agentify_lints
<!-- @on-device:begin -->
  • packages/flutter_agentify_on_device
<!-- @on-device:end -->
  • the skill folders the framework ships in .agents/skills/ (not the generated actions/references/action-reference.md)
  • learnings.defaults.md and tool/upgrade.dart
  • the example folder in .agents/example/, when the project has one

Why

Data changes are what the app is for, and actions already check, record and gate them. Source changes can break layers, tests and the assistant's view of the app, so the gate decides. Dependencies, platform files and CI affect every build and every person who clones the repo, so a person should choose them. Secrets leak by being written once. Generated files are overwritten on the next build, so an edit to one is lost or fights the generator.

How

  • Need to change a generated file? Change its source (the drift table, the annotated provider) and run dart run build_runner build.
  • Need a lint rule off for one line? Put // ignore: flutter_agentify_lints/<rule> on the line above, with a reason. Every // ignore:, ! and dynamic is counted; the allowed totals are in tool/baselines/ratchet.json, and they may only go down. Fix the cause first; write the code differently before you reach for an ignore.
  • analysis_options.yaml lists the enabled lint rules. Turning one off there is a change to the project's rules, so ask first.
  • Need a new package capability? Ask whether the task is about the framework. If it is, work in packages/ and run the package tests too; if it is not, build the feature in lib/ on top of what exists.
  • Keys never go into source, assets, tests, fixtures or prompts. Use the secure storage path described in secrets.
  • Before deleting data on the person's behalf, use the action with its approval step; do not delete database files.

Anti-patterns

  • Editing *.drift.dart by hand to make an error go away.
  • Running dart run flutter_agentify_core:doctor --update-baselines to make a failing ratchet pass. That only records a worse number.
  • Adding a package because it saves ten lines.
  • Adding an Android permission or iOS usage string without asking.
  • Copying a key into a test "just for now".
  • Editing a framework skill to change how the agent behaves, unless the task is to change Flutter Agentify itself; ask the person instead.
  • Editing packages/ code to dodge a lint rule in lib/.

Verification

  • dart run flutter_agentify_core:doctor --only ratchet fails when the counts of // ignore:, ! and dynamic rise above tool/baselines/ratchet.json.
  • dart run flutter_agentify_core:doctor --only bundled-secrets fails on key-shaped text in bundled assets and package source.
  • tool/check.sh runs all of it, plus analysis with the custom lint rules. analysis_options.yaml is not read by the doctor, and no automated check stops you editing a platform folder, CI file or packages/; those boundaries rest on this rule.

Related skills

  • agent-workflow — when to ask, and how to report the change.
  • secrets — where keys live and how they stay out of the repo.
  • upgrade-the-framework — changing framework-owned paths the right way.
  • verify — running the gate and reading its output.
  • actions — changing app data the right way.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.